System capable of establishing model for cardiac ventricular hypertrophy screening

Abstract

A system for establishing a model for cardiac ventricular hypertrophy (VH) screening includes a storage and a processor. The storage stores multiple pieces of subject data respectively associated with multiple subjects. Each of the pieces of subject data contains a basic physiological parameter group, an electrocardiographic parameter group, and an actual VH condition that corresponds to a left or right ventricle of the subject associated with the piece of subject data. The processor is electrically connected to the storage, splits the pieces of subject data into a training set and a test set, and establishes the model for VH screening based on the pieces of subject data in the training set by using machine learning techniques.

What is claimed is: 
     
       1. A system capable of establishing a model for cardiac ventricular hypertrophy (VH) screening, said system comprising:
 a storage configured to store plural pieces of subject data that are respectively associated with a plurality of subjects, each of the pieces of subject data containing a basic physiological parameter group, an electrocardiographic parameter group, and an actual cardiac VH condition that corresponds to a specified ventricle of one of the subjects associated with the piece of subject data, the specified ventricle being one of a left ventricle and a right ventricle; and 
 a processor electrically connected to said storage, and configured to split the pieces of subject data into a training set and a test set, and to establish the model for cardiac VH screening with respect to the specified ventricle based on the pieces of subject data in the training set by using machine learning techniques. 
 
     
     
       2. The system as claimed in  claim 1 , wherein the basic physiological parameter group contains values respectively related to an age, a body height and a body weight of the associated one of the subjects. 
     
     
       3. The system as claimed in  claim 1 , wherein the electrocardiographic parameter group contains values respectively related to a heart rate, a P wave duration in lead II (II-P), a duration of PR segment in lead II (II-PR), a duration of QRS complex in lead II (II-QRS), a duration of QT interval in lead II (II-QT), a duration of corrected QT interval in lead II (II-QTc), a P wave axis in lead II (II-P axis), a QRS axis in lead II (II-QRS axis), a T-wave axis in lead II (II-T axis), an amplitude of R wave in lead I (I-R), an amplitude of R wave in lead II (II-R), an amplitude of R wave in lead III (III-R), an amplitude of R wave in lead augmented vector right (aVR-R), an amplitude of R-wave in lead augmented vector left (aVL-R), an amplitude of S wave in lead augmented vector left (aVL-S), an amplitude of R wave in lead augmented vector foot (aVF-R), an amplitude of R wave in precordial lead V1 (V1-R), an amplitude of S wave in precordial lead V1 (V1-S), an amplitude of R wave in precordial lead V2 (V2-R), an amplitude of S wave in precordial lead V2 (V2-S), an amplitude of R wave in precordial lead V3 (V3-R), an amplitude of S wave in precordial lead V3 (V3-S), an amplitude of R wave in precordial lead V4 (V4-R), an amplitude of S wave in precordial lead V4 (V4 -S), an amplitude of R wave in precordial lead V5 (V5-R), an amplitude of S wave in precordial lead V5 (V5-S), an amplitude of R wave in precordial lead V6 (V6-R) and an amplitude of S wave in precordial lead V6 (V6-S) recorded in an electrocardiogram of the associated one of the subjects. 
     
     
       4. The system as claimed in  claim 1 , wherein said processor is further configured to:
 perform linear transformation on the pieces of subject data by using min-max normalization so as to generate plural pieces of normalized subject data, where the values contained in the basic physiological parameter group and the electrocardiographic parameter group are mapped into a numeric range [0,1] in the normalized subject data; and 
 split the pieces of normalized subject data into the training set and the test set. 
 
     
     
       5. The system as claimed in  claim 1 , wherein said processor is configured to establish, by using a support vector machine (SVM) algorithm, the model for cardiac VH screening based on the pieces of subject data in the training set and a target hyperparameter used in the SVM algorithm. 
     
     
       6. The system as claimed in  claim 5 , wherein said processor is configured to:
 perform data synthesis on the pieces of subject data in the training set to generate plural pieces of synthesized subject data; and 
 establish the model for cardiac VH screening based on the pieces of synthesized subject data and the target hyperparameter by using the SVM algorithm. 
 
     
     
       7. The system as claimed in  claim 6 , wherein said processor is configured to perform data synthesis on the pieces of subject data in the training set by using synthetic minority over-sampling technique (SMOTE) so as to generate synthetic minority class samples, which form a part of the pieces of synthesized subject data, where the pieces of synthesized subject data belonging to a minority class and the pieces of synthesized subject data belonging to a majority class are substantially equal in number. 
     
     
       8. The system as claimed in  claim 5 , wherein the training set includes N training subsets, where N is an integer greater than one; and wherein said processor is configured to:
 select one of the N training subsets as a validation subset, and obtain N-1 other subsets which are all the N training subsets except the validation subset; 
 perform data synthesis on the pieces of subject data in the N-1 other subsets to generate plural pieces of synthetic subject data; and 
 establish, by using the SVM algorithm based on the pieces of synthetic subject data and a plurality of candidate hyperparameters used in the SVM algorithm, a plurality of training models respectively corresponding to the candidate hyperparameters. 
 
     
     
       9. The system as claimed in  claim 8 , wherein said processor is configured to perform data synthesis on the pieces of subject data in the N-1 other subsets by using synthetic minority over-sampling technique (SMOTE) so as to generate synthetic minority class samples, which form a part of the pieces of synthetic subject data, where the pieces of synthetic subject data belonging to a minority class and the pieces of synthetic subject data belonging to a majority class are substantially equal in number. 
     
     
       10. The system as claimed in  claim 5 , wherein the training set includes N training subsets, where N is an integer greater than one, said processor being configured to:
 select one of the N training subsets as a validation subset, and obtain N-1 other subsets which are all the N training subsets except the validation subset; 
 establish, by using the SVM algorithm based on the pieces of subject data in the N-1 other subsets and a plurality of candidate hyperparameters used in the SVM algorithm, a plurality of training models respectively corresponding to the candidate hyperparameters; 
 obtain, based on the training models and the pieces of subject data in the validation subset, a plurality of precision-recall (PR) curves respectively corresponding to the candidate hyperparameters; 
 calculate areas under curves (AUCs) respectively corresponding to the PR curves; and 
 determine, based on the AUCs, one of the candidate hyperparameters to serve as the target hyperparameter. 
 
     
     
       11. The system as claimed in  claim 5 , wherein:
 the training set includes N training subsets, where N is an integer greater than one; 
 said processor is configured to, for each of candidate hyperparameters that are used in the SVM algorithm,
 perform an area under curve (AUC) calculation procedure N number of times so as to obtain N number of AUCs of precision-recall (PR) curves corresponding to the candidate hyperparameter, where the N number of times of performing the AUC calculation procedure respectively corresponding to the N training subsets, and 
 calculate an average AUC value based on the N number of AUCs obtained through the N number of times of performing the AUC calculation procedure; 
 
 said processor is further configured to determine, based on the average AUC values calculated respectively for the candidate hyperparameters, one of the candidate hyperparameters to serve as the target hyperparameter; and 
 wherein said processor is further configured to, in each performance of the AUC calculation procedure for each of the candidate hyperparameters,
 select the corresponding one of the N training subsets as a validation subset, and obtain N-1 other subsets which are all the N training subsets except the validation subset, 
 establish, by using the SVM algorithm based on the candidate hyperparameter and the pieces of subject data in the N-1 other subsets, a training model corresponding to the candidate hyperparameter, 
 obtain, based on the training model and the pieces of subject data in the validation subset, the PR curve corresponding to the candidate hyperparameter, and 
 calculate the AUC for the PR curve corresponding to the candidate hyperparameter.